基于ASTER图像的干旱区土壤盐碱化遥感应用研究

Remote Sensing Application in Studying Soil Salinization in Arid Areas Based on ASTER Images

土壤盐碱化是干旱地区常见的一种土地退化现象。为了对土壤盐碱化程度进行快速的评估，采用ASTER遥感数据敏感波段组合结合光谱角度制图(SAM)的方法进行了探索性的研究。研究从分析干旱区盐碱地表光谱特征的角度出发，采用图像像元纯度指数分析(PPI)方法提取不同盐碱化水平的地表特征纯像元，结合实地测量光谱数据和采样分析数据确定纯像元的物理意义，并应用光谱角度制图方法进行了土壤盐碱化程度分级制图。经检验，总体分类精度达到79．1％。这种方法对常规数据的依赖性较小，提供了一种适合盐碱化快速定量评估的有效途径。

Soil salinity caused by natural or human-induced processes is a major environment hazard and a major cause of soil degradation. Surface salinity is a complex system and a dynamic process, and is affected by soil property, groundwater and terrain. Remote sensing can be used to detect salt-related surface features, and it has been used in soil salinity mapping in recent papers. But it is difficult to get quantitative results of soil salinity from remote sensing application if without plentiful auxiliary data, such as groundwater depth and groundwater mineralization, which are insufficient and difficult to obtain, especially in remote arid areas. In this paper, a remote sensing method for monitoring soil salinization is studied and performed at test site, the Ebinur Lake area located in northwest China, a typical arid environment. The study focuses on assessing extents of soil salinization and mapping soil salinity rapidly, based on analyzing and extracting the information of surface features with classification method of spectral angle mapping （SAM）, a usual means of hyperspectral analysis.To comprehend behavior of salt in soil and get training data, we performed fieldwork about surface features, soil moisture content, vegetation, salt content in soil, etc. To extract the information of salt efficiently, a remote sensing dataset comprised of g-band multi-spectral images of ASTER is used. Supported by synthesizing and analyzing means of GIS, the spectral curves of different extents of salt-affected surfaces are plotted based on the database and the locations of sampling sites. Then, the multi-spectral dataset is performed, and a transformation of minimum noise fraction （MNF） is segregated in the dataset so as to reduce the computational requirements for subsequent processing. The transformed dataset is processed with a pixel purity index and n-dimensional visualizer means to achieve typical salt-affected surface features which are distinguishable spectrally. According to the spectral curves of samples, some pure pixels are gotten, which are some typical salt-affected surface features, distinguishable spectrally, such as the salinized swamps, wet salty crusts, dry puffy salty crusts, sparse vegetations, luxuriant vegetations, eroded land surface and unaffected land surface, which are salinized seriously, moderately, slightly, or unaffected. The reference spectral curves of the features are extracted and used in classification of SAM. Ultimately, soil salinity is mapped in different extents. Assessed with field-derived data, the accuracy of the method is 79.1%.